2 * Copyright 1998 Massachusetts Institute of Technology
3 * Copyright 2012 ADARA Networks, Inc.
4 * Copyright 2017 Dell EMC Isilon
6 * Portions of this software were developed by Robert N. M. Watson under
7 * contract to ADARA Networks, Inc.
9 * Permission to use, copy, modify, and distribute this software and
10 * its documentation for any purpose and without fee is hereby
11 * granted, provided that both the above copyright notice and this
12 * permission notice appear in all copies, that both the above
13 * copyright notice and this permission notice appear in all
14 * supporting documentation, and that the name of M.I.T. not be used
15 * in advertising or publicity pertaining to distribution of the
16 * software without specific, written prior permission. M.I.T. makes
17 * no representations about the suitability of this software for any
18 * purpose. It is provided "as is" without express or implied
21 * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''. M.I.T. DISCLAIMS
22 * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
23 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
24 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
25 * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
28 * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
29 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * if_vlan.c - pseudo-device driver for IEEE 802.1Q virtual LANs.
37 * This is sort of sneaky in the implementation, since
38 * we need to pretend to be enough of an Ethernet implementation
39 * to make arp work. The way we do this is by telling everyone
40 * that we are an Ethernet, and then catch the packets that
41 * ether_output() sends to us via if_transmit(), rewrite them for
42 * use by the real outgoing interface, and ask it to send them.
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
51 #include <sys/param.h>
52 #include <sys/eventhandler.h>
53 #include <sys/kernel.h>
55 #include <sys/malloc.h>
57 #include <sys/module.h>
58 #include <sys/rmlock.h>
60 #include <sys/queue.h>
61 #include <sys/socket.h>
62 #include <sys/sockio.h>
63 #include <sys/sysctl.h>
64 #include <sys/systm.h>
66 #include <sys/taskqueue.h>
69 #include <net/ethernet.h>
71 #include <net/if_var.h>
72 #include <net/if_clone.h>
73 #include <net/if_dl.h>
74 #include <net/if_types.h>
75 #include <net/if_vlan_var.h>
79 #include <netinet/in.h>
80 #include <netinet/if_ether.h>
83 #define VLAN_DEF_HWIDTH 4
84 #define VLAN_IFFLAGS (IFF_BROADCAST | IFF_MULTICAST)
86 #define UP_AND_RUNNING(ifp) \
87 ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
89 LIST_HEAD(ifvlanhead, ifvlan);
92 struct ifnet *parent; /* parent interface of this trunk */
95 #define VLAN_ARRAY_SIZE (EVL_VLID_MASK + 1)
96 struct ifvlan *vlans[VLAN_ARRAY_SIZE]; /* static table */
98 struct ifvlanhead *hash; /* dynamic hash-list table */
106 * This macro provides a facility to iterate over every vlan on a trunk with
107 * the assumption that none will be added/removed during iteration.
110 #define VLAN_FOREACH(_ifv, _trunk) \
112 for (_i = 0; _i < VLAN_ARRAY_SIZE; _i++) \
113 if (((_ifv) = (_trunk)->vlans[_i]) != NULL)
114 #else /* VLAN_ARRAY */
115 #define VLAN_FOREACH(_ifv, _trunk) \
116 struct ifvlan *_next; \
118 for (_i = 0; _i < (1 << (_trunk)->hwidth); _i++) \
119 LIST_FOREACH_SAFE((_ifv), &(_trunk)->hash[_i], ifv_list, _next)
120 #endif /* VLAN_ARRAY */
123 * This macro provides a facility to iterate over every vlan on a trunk while
124 * also modifying the number of vlans on the trunk. The iteration continues
125 * until some condition is met or there are no more vlans on the trunk.
128 /* The VLAN_ARRAY case is simple -- just a for loop using the condition. */
129 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
131 for (_i = 0; !(_cond) && _i < VLAN_ARRAY_SIZE; _i++) \
132 if (((_ifv) = (_trunk)->vlans[_i]))
133 #else /* VLAN_ARRAY */
135 * The hash table case is more complicated. We allow for the hash table to be
136 * modified (i.e. vlans removed) while we are iterating over it. To allow for
137 * this we must restart the iteration every time we "touch" something during
138 * the iteration, since removal will resize the hash table and invalidate our
139 * current position. If acting on the touched element causes the trunk to be
140 * emptied, then iteration also stops.
142 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
144 bool _touch = false; \
146 !(_cond) && _i < (1 << (_trunk)->hwidth); \
147 _i = (_touch && ((_trunk) != NULL) ? 0 : _i + 1), _touch = false) \
148 if (((_ifv) = LIST_FIRST(&(_trunk)->hash[_i])) != NULL && \
150 #endif /* VLAN_ARRAY */
152 struct vlan_mc_entry {
153 struct sockaddr_dl mc_addr;
154 SLIST_ENTRY(vlan_mc_entry) mc_entries;
158 struct ifvlantrunk *ifv_trunk;
159 struct ifnet *ifv_ifp;
160 #define TRUNK(ifv) ((ifv)->ifv_trunk)
161 #define PARENT(ifv) ((ifv)->ifv_trunk->parent)
163 int ifv_pflags; /* special flags we have set on parent */
165 int ifv_encaplen; /* encapsulation length */
166 int ifv_mtufudge; /* MTU fudged by this much */
167 int ifv_mintu; /* min transmission unit */
168 uint16_t ifv_proto; /* encapsulation ethertype */
169 uint16_t ifv_tag; /* tag to apply on packets leaving if */
170 uint16_t ifv_vid; /* VLAN ID */
171 uint8_t ifv_pcp; /* Priority Code Point (PCP). */
172 struct task lladdr_task;
173 SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
175 LIST_ENTRY(ifvlan) ifv_list;
179 /* Special flags we should propagate to parent. */
182 int (*func)(struct ifnet *, int);
184 {IFF_PROMISC, ifpromisc},
185 {IFF_ALLMULTI, if_allmulti},
189 extern int vlan_mtag_pcp;
191 static const char vlanname[] = "vlan";
192 static MALLOC_DEFINE(M_VLAN, vlanname, "802.1Q Virtual LAN Interface");
194 static eventhandler_tag ifdetach_tag;
195 static eventhandler_tag iflladdr_tag;
198 * if_vlan uses two module-level locks to allow concurrent modification of vlan
199 * interfaces and (mostly) allow for vlans to be destroyed while they are being
200 * used for tx/rx. To accomplish this in a way that has acceptable performance
201 * and cooperation with other parts of the network stack there is a
202 * non-sleepable rmlock(9) and an sx(9). Both locks are exclusively acquired
203 * when destroying a vlan interface, i.e. when the if_vlantrunk field of struct
204 * ifnet is de-allocated and NULL'd. Thus a reader holding either lock has a
205 * guarantee that the struct ifvlantrunk references a valid vlan trunk.
207 * The performance-sensitive paths that warrant using the rmlock(9) are
208 * vlan_transmit and vlan_input. Both have to check for the vlan interface's
209 * existence using if_vlantrunk, and being in the network tx/rx paths the use
210 * of an rmlock(9) gives a measureable improvement in performance.
212 * The reason for having an sx(9) is mostly because there are still areas that
213 * must be sleepable and also have safe concurrent access to a vlan interface.
214 * Since the sx(9) exists, it is used by default in most paths unless sleeping
215 * is not permitted, or if it is not clear whether sleeping is permitted.
217 * Note that despite these protections, there is still an inherent race in the
218 * destruction of vlans since there's no guarantee that the ifnet hasn't been
219 * freed/reused when the tx/rx functions are called by the stack. This can only
220 * be fixed by addressing ifnet's lifetime issues.
222 #define _VLAN_RM_ID ifv_rm_lock
223 #define _VLAN_SX_ID ifv_sx
225 static struct rmlock _VLAN_RM_ID;
226 static struct sx _VLAN_SX_ID;
228 #define VLAN_LOCKING_INIT() \
229 rm_init(&_VLAN_RM_ID, "vlan_rm"); \
230 sx_init(&_VLAN_SX_ID, "vlan_sx")
232 #define VLAN_LOCKING_DESTROY() \
233 rm_destroy(&_VLAN_RM_ID); \
234 sx_destroy(&_VLAN_SX_ID)
236 #define _VLAN_RM_TRACKER _vlan_rm_tracker
237 #define VLAN_RLOCK() rm_rlock(&_VLAN_RM_ID, \
239 #define VLAN_RUNLOCK() rm_runlock(&_VLAN_RM_ID, \
241 #define VLAN_WLOCK() rm_wlock(&_VLAN_RM_ID)
242 #define VLAN_WUNLOCK() rm_wunlock(&_VLAN_RM_ID)
243 #define VLAN_RLOCK_ASSERT() rm_assert(&_VLAN_RM_ID, RA_RLOCKED)
244 #define VLAN_WLOCK_ASSERT() rm_assert(&_VLAN_RM_ID, RA_WLOCKED)
245 #define VLAN_RWLOCK_ASSERT() rm_assert(&_VLAN_RM_ID, RA_LOCKED)
246 #define VLAN_LOCK_READER struct rm_priotracker _VLAN_RM_TRACKER
248 #define VLAN_SLOCK() sx_slock(&_VLAN_SX_ID)
249 #define VLAN_SUNLOCK() sx_sunlock(&_VLAN_SX_ID)
250 #define VLAN_XLOCK() sx_xlock(&_VLAN_SX_ID)
251 #define VLAN_XUNLOCK() sx_xunlock(&_VLAN_SX_ID)
252 #define VLAN_SLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_SLOCKED)
253 #define VLAN_XLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_XLOCKED)
254 #define VLAN_SXLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_LOCKED)
258 * We also have a per-trunk rmlock(9), that is locked shared on packet
259 * processing and exclusive when configuration is changed. Note: This should
260 * only be acquired while there is a shared lock on either of the global locks
261 * via VLAN_SLOCK or VLAN_RLOCK. Thus, an exclusive lock on the global locks
262 * makes a call to TRUNK_RLOCK/TRUNK_WLOCK technically superfluous.
264 #define _TRUNK_RM_TRACKER _trunk_rm_tracker
265 #define TRUNK_LOCK_INIT(trunk) rm_init(&(trunk)->lock, vlanname)
266 #define TRUNK_LOCK_DESTROY(trunk) rm_destroy(&(trunk)->lock)
267 #define TRUNK_RLOCK(trunk) rm_rlock(&(trunk)->lock, \
269 #define TRUNK_WLOCK(trunk) rm_wlock(&(trunk)->lock)
270 #define TRUNK_RUNLOCK(trunk) rm_runlock(&(trunk)->lock, \
272 #define TRUNK_WUNLOCK(trunk) rm_wunlock(&(trunk)->lock)
273 #define TRUNK_RLOCK_ASSERT(trunk) rm_assert(&(trunk)->lock, RA_RLOCKED)
274 #define TRUNK_LOCK_ASSERT(trunk) rm_assert(&(trunk)->lock, RA_LOCKED)
275 #define TRUNK_WLOCK_ASSERT(trunk) rm_assert(&(trunk)->lock, RA_WLOCKED)
276 #define TRUNK_LOCK_READER struct rm_priotracker _TRUNK_RM_TRACKER
279 * The VLAN_ARRAY substitutes the dynamic hash with a static array
280 * with 4096 entries. In theory this can give a boost in processing,
281 * however in practice it does not. Probably this is because the array
282 * is too big to fit into CPU cache.
285 static void vlan_inithash(struct ifvlantrunk *trunk);
286 static void vlan_freehash(struct ifvlantrunk *trunk);
287 static int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
288 static int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
289 static void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
290 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
293 static void trunk_destroy(struct ifvlantrunk *trunk);
295 static void vlan_init(void *foo);
296 static void vlan_input(struct ifnet *ifp, struct mbuf *m);
297 static int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
298 static void vlan_qflush(struct ifnet *ifp);
299 static int vlan_setflag(struct ifnet *ifp, int flag, int status,
300 int (*func)(struct ifnet *, int));
301 static int vlan_setflags(struct ifnet *ifp, int status);
302 static int vlan_setmulti(struct ifnet *ifp);
303 static int vlan_transmit(struct ifnet *ifp, struct mbuf *m);
304 static void vlan_unconfig(struct ifnet *ifp);
305 static void vlan_unconfig_locked(struct ifnet *ifp, int departing);
306 static int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
307 static void vlan_link_state(struct ifnet *ifp);
308 static void vlan_capabilities(struct ifvlan *ifv);
309 static void vlan_trunk_capabilities(struct ifnet *ifp);
311 static struct ifnet *vlan_clone_match_ethervid(const char *, int *);
312 static int vlan_clone_match(struct if_clone *, const char *);
313 static int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
314 static int vlan_clone_destroy(struct if_clone *, struct ifnet *);
316 static void vlan_ifdetach(void *arg, struct ifnet *ifp);
317 static void vlan_iflladdr(void *arg, struct ifnet *ifp);
319 static void vlan_lladdr_fn(void *arg, int pending);
321 static struct if_clone *vlan_cloner;
324 static VNET_DEFINE(struct if_clone *, vlan_cloner);
325 #define V_vlan_cloner VNET(vlan_cloner)
329 #define HASH(n, m) ((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
332 vlan_inithash(struct ifvlantrunk *trunk)
337 * The trunk must not be locked here since we call malloc(M_WAITOK).
338 * It is OK in case this function is called before the trunk struct
339 * gets hooked up and becomes visible from other threads.
342 KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
343 ("%s: hash already initialized", __func__));
345 trunk->hwidth = VLAN_DEF_HWIDTH;
346 n = 1 << trunk->hwidth;
347 trunk->hmask = n - 1;
348 trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
349 for (i = 0; i < n; i++)
350 LIST_INIT(&trunk->hash[i]);
354 vlan_freehash(struct ifvlantrunk *trunk)
359 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
360 for (i = 0; i < (1 << trunk->hwidth); i++)
361 KASSERT(LIST_EMPTY(&trunk->hash[i]),
362 ("%s: hash table not empty", __func__));
364 free(trunk->hash, M_VLAN);
366 trunk->hwidth = trunk->hmask = 0;
370 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
375 TRUNK_WLOCK_ASSERT(trunk);
376 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
378 b = 1 << trunk->hwidth;
379 i = HASH(ifv->ifv_vid, trunk->hmask);
380 LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
381 if (ifv->ifv_vid == ifv2->ifv_vid)
385 * Grow the hash when the number of vlans exceeds half of the number of
386 * hash buckets squared. This will make the average linked-list length
389 if (trunk->refcnt > (b * b) / 2) {
390 vlan_growhash(trunk, 1);
391 i = HASH(ifv->ifv_vid, trunk->hmask);
393 LIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
400 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
405 TRUNK_WLOCK_ASSERT(trunk);
406 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
408 b = 1 << trunk->hwidth;
409 i = HASH(ifv->ifv_vid, trunk->hmask);
410 LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
413 LIST_REMOVE(ifv2, ifv_list);
414 if (trunk->refcnt < (b * b) / 2)
415 vlan_growhash(trunk, -1);
419 panic("%s: vlan not found\n", __func__);
420 return (ENOENT); /*NOTREACHED*/
424 * Grow the hash larger or smaller if memory permits.
427 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
430 struct ifvlanhead *hash2;
431 int hwidth2, i, j, n, n2;
433 TRUNK_WLOCK_ASSERT(trunk);
434 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
437 /* Harmless yet obvious coding error */
438 printf("%s: howmuch is 0\n", __func__);
442 hwidth2 = trunk->hwidth + howmuch;
443 n = 1 << trunk->hwidth;
445 /* Do not shrink the table below the default */
446 if (hwidth2 < VLAN_DEF_HWIDTH)
449 /* M_NOWAIT because we're called with trunk mutex held */
450 hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_NOWAIT);
452 printf("%s: out of memory -- hash size not changed\n",
454 return; /* We can live with the old hash table */
456 for (j = 0; j < n2; j++)
457 LIST_INIT(&hash2[j]);
458 for (i = 0; i < n; i++)
459 while ((ifv = LIST_FIRST(&trunk->hash[i])) != NULL) {
460 LIST_REMOVE(ifv, ifv_list);
461 j = HASH(ifv->ifv_vid, n2 - 1);
462 LIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
464 free(trunk->hash, M_VLAN);
466 trunk->hwidth = hwidth2;
467 trunk->hmask = n2 - 1;
470 if_printf(trunk->parent,
471 "VLAN hash table resized from %d to %d buckets\n", n, n2);
474 static __inline struct ifvlan *
475 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
479 TRUNK_RLOCK_ASSERT(trunk);
481 LIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list)
482 if (ifv->ifv_vid == vid)
488 /* Debugging code to view the hashtables. */
490 vlan_dumphash(struct ifvlantrunk *trunk)
495 for (i = 0; i < (1 << trunk->hwidth); i++) {
497 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
498 printf("%s ", ifv->ifv_ifp->if_xname);
505 static __inline struct ifvlan *
506 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
509 return trunk->vlans[vid];
513 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
516 if (trunk->vlans[ifv->ifv_vid] != NULL)
518 trunk->vlans[ifv->ifv_vid] = ifv;
525 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
528 trunk->vlans[ifv->ifv_vid] = NULL;
535 vlan_freehash(struct ifvlantrunk *trunk)
540 vlan_inithash(struct ifvlantrunk *trunk)
544 #endif /* !VLAN_ARRAY */
547 trunk_destroy(struct ifvlantrunk *trunk)
552 vlan_freehash(trunk);
553 trunk->parent->if_vlantrunk = NULL;
554 TRUNK_LOCK_DESTROY(trunk);
555 if_rele(trunk->parent);
560 * Program our multicast filter. What we're actually doing is
561 * programming the multicast filter of the parent. This has the
562 * side effect of causing the parent interface to receive multicast
563 * traffic that it doesn't really want, which ends up being discarded
564 * later by the upper protocol layers. Unfortunately, there's no way
565 * to avoid this: there really is only one physical interface.
568 vlan_setmulti(struct ifnet *ifp)
571 struct ifmultiaddr *ifma;
573 struct vlan_mc_entry *mc;
577 * XXX This stupidly needs the rmlock to avoid sleeping while holding
578 * the in6_multi_mtx (see in6_mc_join_locked).
580 VLAN_RWLOCK_ASSERT();
582 /* Find the parent. */
584 TRUNK_WLOCK_ASSERT(TRUNK(sc));
587 CURVNET_SET_QUIET(ifp_p->if_vnet);
589 /* First, remove any existing filter entries. */
590 while ((mc = SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
591 SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
592 (void)if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
596 /* Now program new ones. */
598 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
599 if (ifma->ifma_addr->sa_family != AF_LINK)
601 mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
603 IF_ADDR_WUNLOCK(ifp);
606 bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
607 mc->mc_addr.sdl_index = ifp_p->if_index;
608 SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
610 IF_ADDR_WUNLOCK(ifp);
611 SLIST_FOREACH (mc, &sc->vlan_mc_listhead, mc_entries) {
612 error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
623 * A handler for parent interface link layer address changes.
624 * If the parent interface link layer address is changed we
625 * should also change it on all children vlans.
628 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
631 struct ifnet *ifv_ifp;
632 struct ifvlantrunk *trunk;
633 struct sockaddr_dl *sdl;
636 /* Need the rmlock since this is run on taskqueue_swi. */
638 trunk = ifp->if_vlantrunk;
645 * OK, it's a trunk. Loop over and change all vlan's lladdrs on it.
646 * We need an exclusive lock here to prevent concurrent SIOCSIFLLADDR
647 * ioctl calls on the parent garbling the lladdr of the child vlan.
650 VLAN_FOREACH(ifv, trunk) {
652 * Copy new new lladdr into the ifv_ifp, enqueue a task
653 * to actually call if_setlladdr. if_setlladdr needs to
654 * be deferred to a taskqueue because it will call into
655 * the if_vlan ioctl path and try to acquire the global
658 ifv_ifp = ifv->ifv_ifp;
659 bcopy(IF_LLADDR(ifp), IF_LLADDR(ifv_ifp),
661 sdl = (struct sockaddr_dl *)ifv_ifp->if_addr->ifa_addr;
662 sdl->sdl_alen = ifp->if_addrlen;
663 taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
665 TRUNK_WUNLOCK(trunk);
670 * A handler for network interface departure events.
671 * Track departure of trunks here so that we don't access invalid
672 * pointers or whatever if a trunk is ripped from under us, e.g.,
673 * by ejecting its hot-plug card. However, if an ifnet is simply
674 * being renamed, then there's no need to tear down the state.
677 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
680 struct ifvlantrunk *trunk;
682 /* If the ifnet is just being renamed, don't do anything. */
683 if (ifp->if_flags & IFF_RENAMING)
686 trunk = ifp->if_vlantrunk;
693 * OK, it's a trunk. Loop over and detach all vlan's on it.
694 * Check trunk pointer after each vlan_unconfig() as it will
695 * free it and set to NULL after the last vlan was detached.
697 VLAN_FOREACH_UNTIL_SAFE(ifv, ifp->if_vlantrunk,
698 ifp->if_vlantrunk == NULL)
699 vlan_unconfig_locked(ifv->ifv_ifp, 1);
701 /* Trunk should have been destroyed in vlan_unconfig(). */
702 KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
707 * Return the trunk device for a virtual interface.
709 static struct ifnet *
710 vlan_trunkdev(struct ifnet *ifp)
715 if (ifp->if_type != IFT_L2VLAN)
718 /* Not clear if callers are sleepable, so acquire the rmlock. */
729 * Return the 12-bit VLAN VID for this interface, for use by external
730 * components such as Infiniband.
732 * XXXRW: Note that the function name here is historical; it should be named
736 vlan_tag(struct ifnet *ifp, uint16_t *vidp)
740 if (ifp->if_type != IFT_L2VLAN)
743 *vidp = ifv->ifv_vid;
748 * Return a driver specific cookie for this interface. Synchronization
749 * with setcookie must be provided by the driver.
752 vlan_cookie(struct ifnet *ifp)
756 if (ifp->if_type != IFT_L2VLAN)
759 return (ifv->ifv_cookie);
763 * Store a cookie in our softc that drivers can use to store driver
764 * private per-instance data in.
767 vlan_setcookie(struct ifnet *ifp, void *cookie)
771 if (ifp->if_type != IFT_L2VLAN)
774 ifv->ifv_cookie = cookie;
779 * Return the vlan device present at the specific VID.
781 static struct ifnet *
782 vlan_devat(struct ifnet *ifp, uint16_t vid)
784 struct ifvlantrunk *trunk;
789 /* Not clear if callers are sleepable, so acquire the rmlock. */
791 trunk = ifp->if_vlantrunk;
798 ifv = vlan_gethash(trunk, vid);
801 TRUNK_RUNLOCK(trunk);
807 * Recalculate the cached VLAN tag exposed via the MIB.
810 vlan_tag_recalculate(struct ifvlan *ifv)
813 ifv->ifv_tag = EVL_MAKETAG(ifv->ifv_vid, ifv->ifv_pcp, 0);
817 * VLAN support can be loaded as a module. The only place in the
818 * system that's intimately aware of this is ether_input. We hook
819 * into this code through vlan_input_p which is defined there and
820 * set here. No one else in the system should be aware of this so
821 * we use an explicit reference here.
823 extern void (*vlan_input_p)(struct ifnet *, struct mbuf *);
825 /* For if_link_state_change() eyes only... */
826 extern void (*vlan_link_state_p)(struct ifnet *);
829 vlan_modevent(module_t mod, int type, void *data)
834 ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
835 vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
836 if (ifdetach_tag == NULL)
838 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
839 vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
840 if (iflladdr_tag == NULL)
843 vlan_input_p = vlan_input;
844 vlan_link_state_p = vlan_link_state;
845 vlan_trunk_cap_p = vlan_trunk_capabilities;
846 vlan_trunkdev_p = vlan_trunkdev;
847 vlan_cookie_p = vlan_cookie;
848 vlan_setcookie_p = vlan_setcookie;
849 vlan_tag_p = vlan_tag;
850 vlan_devat_p = vlan_devat;
852 vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
853 vlan_clone_create, vlan_clone_destroy);
856 printf("vlan: initialized, using "
860 "hash tables with chaining"
867 if_clone_detach(vlan_cloner);
869 EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
870 EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
872 vlan_link_state_p = NULL;
873 vlan_trunk_cap_p = NULL;
874 vlan_trunkdev_p = NULL;
876 vlan_cookie_p = NULL;
877 vlan_setcookie_p = NULL;
879 VLAN_LOCKING_DESTROY();
881 printf("vlan: unloaded\n");
889 static moduledata_t vlan_mod = {
895 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
896 MODULE_VERSION(if_vlan, 3);
900 vnet_vlan_init(const void *unused __unused)
903 vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
904 vlan_clone_create, vlan_clone_destroy);
905 V_vlan_cloner = vlan_cloner;
907 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
908 vnet_vlan_init, NULL);
911 vnet_vlan_uninit(const void *unused __unused)
914 if_clone_detach(V_vlan_cloner);
916 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST,
917 vnet_vlan_uninit, NULL);
921 * Check for <etherif>.<vlan> style interface names.
923 static struct ifnet *
924 vlan_clone_match_ethervid(const char *name, int *vidp)
926 char ifname[IFNAMSIZ];
931 strlcpy(ifname, name, IFNAMSIZ);
932 if ((cp = strchr(ifname, '.')) == NULL)
935 if ((ifp = ifunit_ref(ifname)) == NULL)
943 for(; *cp >= '0' && *cp <= '9'; cp++)
944 vid = (vid * 10) + (*cp - '0');
956 vlan_clone_match(struct if_clone *ifc, const char *name)
960 if (vlan_clone_match_ethervid(name, NULL) != NULL)
963 if (strncmp(vlanname, name, strlen(vlanname)) != 0)
965 for (cp = name + 4; *cp != '\0'; cp++) {
966 if (*cp < '0' || *cp > '9')
974 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
985 struct sockaddr_dl *sdl;
987 static const u_char eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
990 * There are 3 (ugh) ways to specify the cloned device:
991 * o pass a parameter block with the clone request.
992 * o specify parameters in the text of the clone device name
993 * o specify no parameters and get an unattached device that
994 * must be configured separately.
995 * The first technique is preferred; the latter two are
996 * supported for backwards compatibility.
998 * XXXRW: Note historic use of the word "tag" here. New ioctls may be
1002 error = copyin(params, &vlr, sizeof(vlr));
1005 p = ifunit_ref(vlr.vlr_parent);
1008 error = ifc_name2unit(name, &unit);
1014 wildcard = (unit < 0);
1015 } else if ((p = vlan_clone_match_ethervid(name, &vid)) != NULL) {
1020 error = ifc_name2unit(name, &unit);
1024 wildcard = (unit < 0);
1027 error = ifc_alloc_unit(ifc, &unit);
1034 /* In the wildcard case, we need to update the name. */
1036 for (dp = name; *dp != '\0'; dp++);
1037 if (snprintf(dp, len - (dp-name), "%d", unit) >
1038 len - (dp-name) - 1) {
1039 panic("%s: interface name too long", __func__);
1043 ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
1044 ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
1046 ifc_free_unit(ifc, unit);
1052 SLIST_INIT(&ifv->vlan_mc_listhead);
1053 ifp->if_softc = ifv;
1055 * Set the name manually rather than using if_initname because
1056 * we don't conform to the default naming convention for interfaces.
1058 strlcpy(ifp->if_xname, name, IFNAMSIZ);
1059 ifp->if_dname = vlanname;
1060 ifp->if_dunit = unit;
1062 ifp->if_init = vlan_init;
1063 ifp->if_transmit = vlan_transmit;
1064 ifp->if_qflush = vlan_qflush;
1065 ifp->if_ioctl = vlan_ioctl;
1066 ifp->if_flags = VLAN_IFFLAGS;
1067 ether_ifattach(ifp, eaddr);
1068 /* Now undo some of the damage... */
1069 ifp->if_baudrate = 0;
1070 ifp->if_type = IFT_L2VLAN;
1071 ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
1073 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1074 sdl->sdl_type = IFT_L2VLAN;
1077 error = vlan_config(ifv, p, vid);
1081 * Since we've partially failed, we need to back
1082 * out all the way, otherwise userland could get
1083 * confused. Thus, we destroy the interface.
1085 ether_ifdetach(ifp);
1088 ifc_free_unit(ifc, unit);
1099 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
1101 struct ifvlan *ifv = ifp->if_softc;
1102 int unit = ifp->if_dunit;
1104 ether_ifdetach(ifp); /* first, remove it from system-wide lists */
1105 vlan_unconfig(ifp); /* now it can be unconfigured and freed */
1107 * We should have the only reference to the ifv now, so we can now
1108 * drain any remaining lladdr task before freeing the ifnet and the
1111 taskqueue_drain(taskqueue_thread, &ifv->lladdr_task);
1114 ifc_free_unit(ifc, unit);
1120 * The ifp->if_init entry point for vlan(4) is a no-op.
1123 vlan_init(void *foo __unused)
1128 * The if_transmit method for vlan(4) interface.
1131 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1135 int error, len, mcast;
1139 ifv = ifp->if_softc;
1140 if (TRUNK(ifv) == NULL) {
1141 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1147 len = m->m_pkthdr.len;
1148 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1153 * Do not run parent's if_transmit() if the parent is not up,
1154 * or parent's driver will cause a system crash.
1156 if (!UP_AND_RUNNING(p)) {
1157 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1163 if (!ether_8021q_frame(&m, ifp, p, ifv->ifv_vid, ifv->ifv_pcp)) {
1164 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1170 * Send it, precisely as ether_output() would have.
1172 error = (p->if_transmit)(p, m);
1174 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1175 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
1176 if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast);
1178 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1184 * The ifp->if_qflush entry point for vlan(4) is a no-op.
1187 vlan_qflush(struct ifnet *ifp __unused)
1192 vlan_input(struct ifnet *ifp, struct mbuf *m)
1194 struct ifvlantrunk *trunk;
1202 trunk = ifp->if_vlantrunk;
1203 if (trunk == NULL) {
1209 if (m->m_flags & M_VLANTAG) {
1211 * Packet is tagged, but m contains a normal
1212 * Ethernet frame; the tag is stored out-of-band.
1214 tag = m->m_pkthdr.ether_vtag;
1215 m->m_flags &= ~M_VLANTAG;
1217 struct ether_vlan_header *evl;
1220 * Packet is tagged in-band as specified by 802.1q.
1222 switch (ifp->if_type) {
1224 if (m->m_len < sizeof(*evl) &&
1225 (m = m_pullup(m, sizeof(*evl))) == NULL) {
1226 if_printf(ifp, "cannot pullup VLAN header\n");
1230 evl = mtod(m, struct ether_vlan_header *);
1231 tag = ntohs(evl->evl_tag);
1234 * Remove the 802.1q header by copying the Ethernet
1235 * addresses over it and adjusting the beginning of
1236 * the data in the mbuf. The encapsulated Ethernet
1237 * type field is already in place.
1239 bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1240 ETHER_HDR_LEN - ETHER_TYPE_LEN);
1241 m_adj(m, ETHER_VLAN_ENCAP_LEN);
1246 panic("%s: %s has unsupported if_type %u",
1247 __func__, ifp->if_xname, ifp->if_type);
1249 if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1256 vid = EVL_VLANOFTAG(tag);
1259 ifv = vlan_gethash(trunk, vid);
1260 if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1261 TRUNK_RUNLOCK(trunk);
1262 if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1267 TRUNK_RUNLOCK(trunk);
1269 if (vlan_mtag_pcp) {
1271 * While uncommon, it is possible that we will find a 802.1q
1272 * packet encapsulated inside another packet that also had an
1273 * 802.1q header. For example, ethernet tunneled over IPSEC
1274 * arriving over ethernet. In that case, we replace the
1275 * existing 802.1q PCP m_tag value.
1277 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
1279 mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_IN,
1280 sizeof(uint8_t), M_NOWAIT);
1282 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1287 m_tag_prepend(m, mtag);
1289 *(uint8_t *)(mtag + 1) = EVL_PRIOFTAG(tag);
1292 m->m_pkthdr.rcvif = ifv->ifv_ifp;
1293 if_inc_counter(ifv->ifv_ifp, IFCOUNTER_IPACKETS, 1);
1296 /* Pass it back through the parent's input routine. */
1297 (*ifv->ifv_ifp->if_input)(ifv->ifv_ifp, m);
1301 vlan_lladdr_fn(void *arg, int pending __unused)
1306 ifv = (struct ifvlan *)arg;
1309 CURVNET_SET(ifp->if_vnet);
1311 /* The ifv_ifp already has the lladdr copied in. */
1312 if_setlladdr(ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1318 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid)
1320 struct ifvlantrunk *trunk;
1325 * We can handle non-ethernet hardware types as long as
1326 * they handle the tagging and headers themselves.
1328 if (p->if_type != IFT_ETHER &&
1329 (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1330 return (EPROTONOSUPPORT);
1331 if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1332 return (EPROTONOSUPPORT);
1334 * Don't let the caller set up a VLAN VID with
1335 * anything except VLID bits.
1336 * VID numbers 0x0 and 0xFFF are reserved.
1338 if (vid == 0 || vid == 0xFFF || (vid & ~EVL_VLID_MASK))
1343 /* Acquire rmlock after the branch so we can M_WAITOK. */
1345 if (p->if_vlantrunk == NULL) {
1346 trunk = malloc(sizeof(struct ifvlantrunk),
1347 M_VLAN, M_WAITOK | M_ZERO);
1348 vlan_inithash(trunk);
1349 TRUNK_LOCK_INIT(trunk);
1352 p->if_vlantrunk = trunk;
1354 if_ref(trunk->parent);
1357 trunk = p->if_vlantrunk;
1361 ifv->ifv_vid = vid; /* must set this before vlan_inshash() */
1362 ifv->ifv_pcp = 0; /* Default: best effort delivery. */
1363 vlan_tag_recalculate(ifv);
1364 error = vlan_inshash(trunk, ifv);
1367 ifv->ifv_proto = ETHERTYPE_VLAN;
1368 ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1369 ifv->ifv_mintu = ETHERMIN;
1370 ifv->ifv_pflags = 0;
1371 ifv->ifv_capenable = -1;
1374 * If the parent supports the VLAN_MTU capability,
1375 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1378 if (p->if_capenable & IFCAP_VLAN_MTU) {
1380 * No need to fudge the MTU since the parent can
1381 * handle extended frames.
1383 ifv->ifv_mtufudge = 0;
1386 * Fudge the MTU by the encapsulation size. This
1387 * makes us incompatible with strictly compliant
1388 * 802.1Q implementations, but allows us to use
1389 * the feature with other NetBSD implementations,
1390 * which might still be useful.
1392 ifv->ifv_mtufudge = ifv->ifv_encaplen;
1395 ifv->ifv_trunk = trunk;
1398 * Initialize fields from our parent. This duplicates some
1399 * work with ether_ifattach() but allows for non-ethernet
1400 * interfaces to also work.
1402 ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1403 ifp->if_baudrate = p->if_baudrate;
1404 ifp->if_output = p->if_output;
1405 ifp->if_input = p->if_input;
1406 ifp->if_resolvemulti = p->if_resolvemulti;
1407 ifp->if_addrlen = p->if_addrlen;
1408 ifp->if_broadcastaddr = p->if_broadcastaddr;
1411 * Copy only a selected subset of flags from the parent.
1412 * Other flags are none of our business.
1414 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1415 ifp->if_flags &= ~VLAN_COPY_FLAGS;
1416 ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1417 #undef VLAN_COPY_FLAGS
1419 ifp->if_link_state = p->if_link_state;
1421 vlan_capabilities(ifv);
1424 * Set up our interface address to reflect the underlying
1425 * physical interface's.
1427 bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1428 ((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1432 * Configure multicast addresses that may already be
1433 * joined on the vlan device.
1435 (void)vlan_setmulti(ifp);
1437 TASK_INIT(&ifv->lladdr_task, 0, vlan_lladdr_fn, ifv);
1439 /* We are ready for operation now. */
1440 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1442 /* Update flags on the parent, if necessary. */
1443 vlan_setflags(ifp, 1);
1446 * We need to drop the non-sleepable rmlock so that the underlying
1447 * devices can sleep in their vlan_config hooks.
1449 TRUNK_WUNLOCK(trunk);
1452 EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid);
1459 vlan_unconfig(struct ifnet *ifp)
1463 vlan_unconfig_locked(ifp, 0);
1468 vlan_unconfig_locked(struct ifnet *ifp, int departing)
1470 struct ifvlantrunk *trunk;
1471 struct vlan_mc_entry *mc;
1473 struct ifnet *parent;
1476 VLAN_XLOCK_ASSERT();
1478 ifv = ifp->if_softc;
1479 trunk = ifv->ifv_trunk;
1482 if (trunk != NULL) {
1484 * Both vlan_transmit and vlan_input rely on the trunk fields
1485 * being NULL to determine whether to bail, so we need to get
1486 * an exclusive lock here to prevent them from using bad
1490 parent = trunk->parent;
1493 * Since the interface is being unconfigured, we need to
1494 * empty the list of multicast groups that we may have joined
1495 * while we were alive from the parent's list.
1497 while ((mc = SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1499 * If the parent interface is being detached,
1500 * all its multicast addresses have already
1501 * been removed. Warn about errors if
1502 * if_delmulti() does fail, but don't abort as
1503 * all callers expect vlan destruction to
1507 error = if_delmulti(parent,
1508 (struct sockaddr *)&mc->mc_addr);
1511 "Failed to delete multicast address from parent: %d\n",
1514 SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1518 vlan_setflags(ifp, 0); /* clear special flags on parent */
1521 * The trunk lock isn't actually required here, but
1522 * vlan_remhash expects it.
1525 vlan_remhash(trunk, ifv);
1526 TRUNK_WUNLOCK(trunk);
1527 ifv->ifv_trunk = NULL;
1530 * Check if we were the last.
1532 if (trunk->refcnt == 0) {
1533 parent->if_vlantrunk = NULL;
1534 trunk_destroy(trunk);
1539 /* Disconnect from parent. */
1540 if (ifv->ifv_pflags)
1541 if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1542 ifp->if_mtu = ETHERMTU;
1543 ifp->if_link_state = LINK_STATE_UNKNOWN;
1544 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1547 * Only dispatch an event if vlan was
1548 * attached, otherwise there is nothing
1549 * to cleanup anyway.
1552 EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid);
1555 /* Handle a reference counted flag that should be set on the parent as well */
1557 vlan_setflag(struct ifnet *ifp, int flag, int status,
1558 int (*func)(struct ifnet *, int))
1563 VLAN_SXLOCK_ASSERT();
1565 ifv = ifp->if_softc;
1566 status = status ? (ifp->if_flags & flag) : 0;
1567 /* Now "status" contains the flag value or 0 */
1570 * See if recorded parent's status is different from what
1571 * we want it to be. If it is, flip it. We record parent's
1572 * status in ifv_pflags so that we won't clear parent's flag
1573 * we haven't set. In fact, we don't clear or set parent's
1574 * flags directly, but get or release references to them.
1575 * That's why we can be sure that recorded flags still are
1576 * in accord with actual parent's flags.
1578 if (status != (ifv->ifv_pflags & flag)) {
1579 error = (*func)(PARENT(ifv), status);
1582 ifv->ifv_pflags &= ~flag;
1583 ifv->ifv_pflags |= status;
1589 * Handle IFF_* flags that require certain changes on the parent:
1590 * if "status" is true, update parent's flags respective to our if_flags;
1591 * if "status" is false, forcedly clear the flags set on parent.
1594 vlan_setflags(struct ifnet *ifp, int status)
1598 for (i = 0; vlan_pflags[i].flag; i++) {
1599 error = vlan_setflag(ifp, vlan_pflags[i].flag,
1600 status, vlan_pflags[i].func);
1607 /* Inform all vlans that their parent has changed link state */
1609 vlan_link_state(struct ifnet *ifp)
1611 struct ifvlantrunk *trunk;
1615 /* Called from a taskqueue_swi task, so we cannot sleep. */
1617 trunk = ifp->if_vlantrunk;
1618 if (trunk == NULL) {
1624 VLAN_FOREACH(ifv, trunk) {
1625 ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1626 if_link_state_change(ifv->ifv_ifp,
1627 trunk->parent->if_link_state);
1629 TRUNK_WUNLOCK(trunk);
1634 vlan_capabilities(struct ifvlan *ifv)
1638 struct ifnet_hw_tsomax hw_tsomax;
1639 int cap = 0, ena = 0, mena;
1642 VLAN_SXLOCK_ASSERT();
1643 TRUNK_WLOCK_ASSERT(TRUNK(ifv));
1647 /* Mask parent interface enabled capabilities disabled by user. */
1648 mena = p->if_capenable & ifv->ifv_capenable;
1651 * If the parent interface can do checksum offloading
1652 * on VLANs, then propagate its hardware-assisted
1653 * checksumming flags. Also assert that checksum
1654 * offloading requires hardware VLAN tagging.
1656 if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1657 cap |= p->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1658 if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1659 p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1660 ena |= mena & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1661 if (ena & IFCAP_TXCSUM)
1662 hwa |= p->if_hwassist & (CSUM_IP | CSUM_TCP |
1663 CSUM_UDP | CSUM_SCTP);
1664 if (ena & IFCAP_TXCSUM_IPV6)
1665 hwa |= p->if_hwassist & (CSUM_TCP_IPV6 |
1666 CSUM_UDP_IPV6 | CSUM_SCTP_IPV6);
1670 * If the parent interface can do TSO on VLANs then
1671 * propagate the hardware-assisted flag. TSO on VLANs
1672 * does not necessarily require hardware VLAN tagging.
1674 memset(&hw_tsomax, 0, sizeof(hw_tsomax));
1675 if_hw_tsomax_common(p, &hw_tsomax);
1676 if_hw_tsomax_update(ifp, &hw_tsomax);
1677 if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1678 cap |= p->if_capabilities & IFCAP_TSO;
1679 if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1680 ena |= mena & IFCAP_TSO;
1681 if (ena & IFCAP_TSO)
1682 hwa |= p->if_hwassist & CSUM_TSO;
1686 * If the parent interface can do LRO and checksum offloading on
1687 * VLANs, then guess it may do LRO on VLANs. False positive here
1688 * cost nothing, while false negative may lead to some confusions.
1690 if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1691 cap |= p->if_capabilities & IFCAP_LRO;
1692 if (p->if_capenable & IFCAP_VLAN_HWCSUM)
1693 ena |= p->if_capenable & IFCAP_LRO;
1696 * If the parent interface can offload TCP connections over VLANs then
1697 * propagate its TOE capability to the VLAN interface.
1699 * All TOE drivers in the tree today can deal with VLANs. If this
1700 * changes then IFCAP_VLAN_TOE should be promoted to a full capability
1703 #define IFCAP_VLAN_TOE IFCAP_TOE
1704 if (p->if_capabilities & IFCAP_VLAN_TOE)
1705 cap |= p->if_capabilities & IFCAP_TOE;
1706 if (p->if_capenable & IFCAP_VLAN_TOE) {
1707 TOEDEV(ifp) = TOEDEV(p);
1708 ena |= mena & IFCAP_TOE;
1712 * If the parent interface supports dynamic link state, so does the
1715 cap |= (p->if_capabilities & IFCAP_LINKSTATE);
1716 ena |= (mena & IFCAP_LINKSTATE);
1718 ifp->if_capabilities = cap;
1719 ifp->if_capenable = ena;
1720 ifp->if_hwassist = hwa;
1724 vlan_trunk_capabilities(struct ifnet *ifp)
1726 struct ifvlantrunk *trunk;
1730 trunk = ifp->if_vlantrunk;
1731 if (trunk == NULL) {
1736 VLAN_FOREACH(ifv, trunk) {
1737 vlan_capabilities(ifv);
1739 TRUNK_WUNLOCK(trunk);
1744 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1750 struct ifvlantrunk *trunk;
1755 ifr = (struct ifreq *)data;
1756 ifa = (struct ifaddr *) data;
1757 ifv = ifp->if_softc;
1761 ifp->if_flags |= IFF_UP;
1763 if (ifa->ifa_addr->sa_family == AF_INET)
1764 arp_ifinit(ifp, ifa);
1768 bcopy(IF_LLADDR(ifp), &ifr->ifr_addr.sa_data[0],
1773 if (TRUNK(ifv) != NULL) {
1776 error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1778 /* Limit the result to the parent's current config. */
1780 struct ifmediareq *ifmr;
1782 ifmr = (struct ifmediareq *)data;
1783 if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1784 ifmr->ifm_count = 1;
1785 error = copyout(&ifmr->ifm_current,
1802 * Set the interface MTU.
1806 if (trunk != NULL) {
1809 (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1811 (ifv->ifv_mintu - ifv->ifv_mtufudge))
1814 ifp->if_mtu = ifr->ifr_mtu;
1815 TRUNK_WUNLOCK(trunk);
1824 * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN
1825 * interface to be delegated to a jail without allowing the
1826 * jail to change what underlying interface/VID it is
1827 * associated with. We are not entirely convinced that this
1828 * is the right way to accomplish that policy goal.
1830 if (ifp->if_vnet != ifp->if_home_vnet) {
1835 error = copyin(ifr_data_get_ptr(ifr), &vlr, sizeof(vlr));
1838 if (vlr.vlr_parent[0] == '\0') {
1842 p = ifunit_ref(vlr.vlr_parent);
1847 error = vlan_config(ifv, p, vlr.vlr_tag);
1853 if (ifp->if_vnet != ifp->if_home_vnet) {
1858 bzero(&vlr, sizeof(vlr));
1860 if (TRUNK(ifv) != NULL) {
1861 strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1862 sizeof(vlr.vlr_parent));
1863 vlr.vlr_tag = ifv->ifv_vid;
1866 error = copyout(&vlr, ifr_data_get_ptr(ifr), sizeof(vlr));
1871 * We should propagate selected flags to the parent,
1872 * e.g., promiscuous mode.
1875 if (TRUNK(ifv) != NULL)
1876 error = vlan_setflags(ifp, 1);
1883 * If we don't have a parent, just remember the membership for
1886 * XXX We need the rmlock here to avoid sleeping while
1887 * holding in6_multi_mtx.
1891 if (trunk != NULL) {
1893 error = vlan_setmulti(ifp);
1894 TRUNK_WUNLOCK(trunk);
1901 if (ifp->if_vnet != ifp->if_home_vnet) {
1906 ifr->ifr_vlan_pcp = ifv->ifv_pcp;
1911 if (ifp->if_vnet != ifp->if_home_vnet) {
1916 error = priv_check(curthread, PRIV_NET_SETVLANPCP);
1919 if (ifr->ifr_vlan_pcp > 7) {
1923 ifv->ifv_pcp = ifr->ifr_vlan_pcp;
1924 vlan_tag_recalculate(ifv);
1925 /* broadcast event about PCP change */
1926 EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_PCP);
1931 ifv->ifv_capenable = ifr->ifr_reqcap;
1933 if (trunk != NULL) {
1935 vlan_capabilities(ifv);
1936 TRUNK_WUNLOCK(trunk);